Registro de imagens tridimensionais das vias aéreas superiores para geração de modelos para pacientes específicos

Abstract

In medical applications, image registration is needed each time two images are being fused in order to increase the resolution of volumetric structures and to facilitate boundaries finding. This work discusses techniques used to register medical images. This work is part of OPAL pro ject. OPAL is a pro ject for dynamic modeling of the oral, pharyngeal, and laryngeal complex for biomedical applications. Part of the research plan for OPAL is to investigate obstructive sleep apnea, a medical condition that causes excessive daytime sleepiness, using Artisynth, a 3D biomechanical modeling toolkit. Patient specific modeling can be divided into four tasks: acquisition, registration, segmentation and morphing. The registration consists in finding the correspondence between two coordinates systems by computing the best rigid body transformation. The volume resolution increasing in MRI images is resulting from the combination of images with directions and slicing different resolutions. To combine these images, monomodal registration techniques are applied to calculate the transformation that aligns the images from the same modality. On the other hand, multimodal registration techniques based on segmentation align images with different information to enhance the tongue contours. In multimodal image registration, CT scans were aligned with MRI scans to combine soft and hard tissues information. The MRI scans contain information about soft tissues such as muscles and CT scans contain about information of the dental cast or dense tissues. A voxel intensity based method is used to determine the rigid-body transformation which aligns MRI scans with different slicing directions by maximization of mutual information. It increases the resolution of the volumetric structure by merging aligned scans. Improving tongue and palate segmentation requires dental cast scan alignment with upper airway merged scans by applying point based techniques for multimodality registration, such as least squares (LS - Least Squares), iterative closest points (ICP - Iterative closest point) and principal components analysis (PCA - Principal Component Analysis). The evaluation is carried out by visual inspection through checkerboard images and distance histograms.